Abstract 15366: Identifying Factors Associated With In-Hospital Mortality in Patients With Takatsubo Cardiomyopathy Utilizing a Machine Learning Algorithm

Abstract

Introduction: Takatsubo cardiomyopathy (TTS), also known as stress induced cardiomyopathy has a reported in-hospital mortality as high as 5% in the US. While overall risk factors have been previously reported, data regarding in-hospital contributors of mortality in TTS is lacking. In our study, we have utilized a Machine Learning (ML) algorithm to identify predictors of in-hospital mortality in patients with TTS. Hypothesis: Machine learning algorithms may be used to identify factors associated with in-hospital mortality in patients with TTS. Methods: We identified all unweighted hospitalizations with a primary diagnosis of TTS from 2016-2019 using International Classification of Diseases, Tenth Revision, diagnosis codes from the National Inpatient Sample (NIS). This is the largest publicly available database of US hospitalizations comprising a 20% stratified sample of all discharges. Data regarding demographics, hospital characteristics and associated diagnoses were abstracted and analyzed using a decision-tree ML algorithm. Results: We identified 7,517 patients who were hospitalized with a primary diagnosis of TTS. In this cohort, in-hospital mortality was noted to be 1.5%. 4 branches, 5 leaves and 8 nodes were identified by the decision-tree algorithm. In our model, the identified covariates associated with high in-hospital mortality were sudden cardiac arrest (p<0.001), cardiogenic shock (p<0.001), sepsis (p<0.001) and Charlson Comorbidity burden (p<0.001) in hierarchical order. Conclusions: ML based algorithms are being increasingly used to analyze clinical data. Our decision-tree ML algorithm was able to identify sudden cardiac arrest, cardiogenic shock, sepsis and Charlson comorbidity burden to be statistically significant predictors of in-hospital mortality amongst 77 other covariates.